创建使用Multiprocessing和Multiprocessing.Queues

时间:2016-02-09 13:28:21

标签: python linux multiprocessing daemon

我有一个任务是监听UDP数据报,解码它们(数据报有二进制信息),解码信息放在字典中,转储字典转换为json字符串,然后将json字符串发送到远程服务器(ActiveMQ)。

解码和发送到远程都可能非常耗时。为了使程序更具可伸缩性,我们创建了两个进程(Multiprocessing.Process):

  • Listner(监听数据报,分析,创建json并将其放入Multiprocessing.Queue中)
  • 发件人(不断尝试从队列中获取json字符串到数组,如果数组长度超过阈值 - 将所有收集的字符串发送到远程服务器)

现在我需要从中创建一个合适的Linux守护进程(可以通过 service 命令启动,停止重启)。 / p>

问题:如何从python多处理程序制作一个守护进程。我没有找到关于此的指南。有人知道怎么做,或者有工作的例子。

以下文字是我的努力来解决这个问题: 我发现了一个python守护进程的小例子:http://www.gavinj.net/2012/06/building-python-daemon-process.html 所以我重写了我的代码(对不起大代码):

import socket
import time
import os    
from select import select    
import multiprocessing
from multiprocessing import Process, Queue, Value

import stomp
import json

import logging
logger = logging.getLogger("DaemonLog")
logger.setLevel(logging.INFO)
formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
handler = logging.FileHandler("/var/log/testdaemon/testdaemon.log")
handler.setFormatter(formatter)
logger.addHandler(handler)
log = logger
#Config listner
domain = 'example.host.ru'
port = int(9930)

#Config remote queue access
queue_cfg = {
    'host': 'queue.test.ru',
    'port': 61113,
    'user': 'user',
    'password': 'pass',
    'queue': '/topic/test.queue'
}

class UDPListener():
    def __init__(self, domain, port, queue_cfg):
        # If I initialize socket during init I see strange error:
        # on the line: data, addr = sock_inst.recvfrom(int(10000))
        # error: [Errno 88] Socket operation on non-socket
        # So I put initialization to runListner function
        #self.sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        #self.sock.bind((domain, port))
        self.domain = domain
        self.port = port
        self.remote_queue_cfg = queue_cfg
        self.queue = Queue()
        self.isWorking = Value('b', True)
        self.decoder = Decoder()
        self.reactor = ParallelQueueReactor(self.queue)

        self.stdin_path = '/dev/null'
        self.stdout_path = '/dev/tty'
        self.stderr_path = '/dev/tty'
        self.pidfile_path = '/var/run/testdaemon/testdaemon.pid'
        self.pidfile_timeout = 5

    def __assignData(self, addr, data):
        receive_time = time.time()
        messages = self.decoder.decode(receive_time, addr, data)
        for msg in messages:
            self.reactor.addMessage(msg)

    def runListner(self):
        self.sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        self.sock.bind((domain, port))
        while self.isWorking.value:
            inputready, outputready, exceptready = select([self.sock], [], [])
            for sock_inst in inputready:
                if sock_inst == self.sock:
                    data, addr = sock_inst.recvfrom(int(10000))
                if data:
                    self.__assignData(addr[0], data)
        self.sock.close()

    def runQueueDispatcher(self):
        while self.isWorking.value:
            connected = False
            while not connected:
                try:
                    conn = stomp.Connection(host_and_ports=[(self.remote_queue_cfg['host'], self.remote_queue_cfg['port'])])
                    conn.start()
                    conn.connect(self.remote_queue_cfg['user'], self.remote_queue_cfg['password'], wait=True)
                    connected = True
                except socket.error:
                    log.error('Could not connect to activemq server.')
                    time.sleep(20)

                if connected == True:
                    while self.isWorking.value:
                        msg = None
                        if not self.queue.empty():
                            #Now error appear hear even when not self.queue.empty()
                            msg = self.queue.get()
                        else:
                            time.sleep(1)

                        if msg is not None:
                            try:
                                data = json.dumps(msg)
                                conn.send(body=data, destination=self.remote_queue_cfg['queue'])
                                count += 1
                            except:
                                log.error('Failed to send message to queue.')
                                time.sleep(1)

    def stop(self):
        self.isWorking.value = False

    def run(self):
        log.error('StartProcesses')
        dispatcher_process = Process(target=self.runQueueDispatcher, name='Dispatcher')
        listner_process = Process(target=self.runListner, name='Listner')
        dispatcher_process.start()
        listner_process.start()
        dispatcher_process.join()
        listner_process.join()
        log.info('Finished')
#------------------------------------------------------------------
def main():
    from daemon import runner

    app = UDPListener(domain, port, queue_cfg)

    daemon_runner = runner.DaemonRunner(app)
    daemon_runner.daemon_context.files_preserve=[handler.stream]
    daemon_runner.do_action()

if __name__ == "__main__":
        main()

现在我在msg = self.queue.get()

上看到错误
Traceback (most recent call last):   File "/usr/lib64/python2.6/multiprocessing/process.py", line 232, in
_bootstrap
    self.run()   File "/usr/lib64/python2.6/multiprocessing/process.py", line 88, in run
    self._target(*self._args, **self._kwargs)   File "/root/ipelevan/dream/src/parallel_main.py", line 116, in runQueueDispatcher
    msg = self.queue.get()   File "/usr/lib64/python2.6/multiprocessing/queues.py", line 91, in get
    res = self._recv() EOFError

手动运行UDPListner.run()时,我没有看到这个错误。但是使用daemon runner看起来UDPListner的新实例在底层和不同的进程中创建,我们有不同的Queue(当它在 init 中初始化时,也有不同的self.socket)。

1 个答案:

答案 0 :(得分:0)

首先:为了按进程使用目的,将共享对象(Queue,Value)的链接保持为类的成员是一个坏主意。它在某种程度上没有妖魔化。但是当在DaemonContext中运行相同的代码时,os.fork()发生了,并以某种方式搞砸了对象的链接。我不太确定Multiprocessing模块是否设计为在对象的方法内100%正确工作。

第二:DaemonContext有助于从shell分离进程,重定向流并执行与守护程序进程相关的其他一些事情但我没有找到任何好的方法来检查这样的守护进程是否已经在运行。所以我只是用了

if os.path.isfile(pidfile_path):
        print 'pidfile %s exists. Already running?' % pidfile_path
        sys.exit(1)